Genetic Hybridization (Hybridgen): A Cooperative Coevolution Algorithm for Variable Selection
Sprache des Titels:
Proceedings of the CAC 2014 conference
HybridGen has been developed following the concept of cooperative coevolution . Its goal is to combine, by means of genetic operators, the searching power of the pure entities, coming from different metaheuristics, leading to hybrids which are better than their predecessors. A prerequisite is isomorphic coding, thus entities can be compared and combined. The general idea is a cyclic process consisting on i) letting each metaheuristic run on a subpopulation, ii) coevolve all the entities together as a whole big population using the genetic operators, and iii) split the coevolved population onto subpopulations, going back to i). We act on all three steps, achieving better solutions with faster convergence and escaping from local minima, by i) controlling each metaheuristic separately, ii) experimenting with the genetic operators during the overall process, and iii) modifying the way the subpopulations are obtained from the joint general population.